OpenAI's Astra becomes the main target for Chinese AI labs after Anthropic tightens its defenses
decoder ·
Chinese artificial intelligence laboratories have shifted their primary focus from Anthropic's Claude to OpenAI's GPT-6 Astra model for training purposes. Reports indicate that companies such as DeepSeek and Moonshot AI are utilizing Astra's outputs to develop new models, including Kimi K3.1 Pro and V4.1-Flash 1017. This distillation technique requires the target model to reveal its hidden reasoning steps, which both OpenAI and Anthropic explicitly prohibit. Anthropic has recently strengthened its security measures, claiming that these new defenses reduce the amount of extractable data by up to 99 percent. Consequently, Chinese labs are targeting Astra because it remains more vulnerable compared to the hardened Claude system. OpenAI previously identified a coordinated distillation campaign involving individuals linked to Moonshot in late September. While OpenAI takes these threats seriously, attackers have historically adapted their tactics whenever new defensive measures were implemented.
decoder